kernel module build error in kernel-2.6.24-rc2-git6

[sangu]

$ rpmbuild --rebuild nvidia-kmod-169.04-1.fc9.src.rpm --define "kernels 2.6.24-0.38.rc2.git6.fc9" --target i686
[...]
make -f /usr/src/kernels/2.6.24-0.38.rc2.git6.fc9-i686/scripts/Makefile.modpost
scripts/mod/modpost -i /usr/src/kernels/2.6.24-0.38.rc2.git6.fc9-i686/Module.symvers -I /usr/src/redhat/BUILD/nvidia-kmod-169.04/_kmod_build_2.6.24-0.38.rc2.git6.fc9/usr/src/nv/Module.symvers -o /usr/src/redhat/BUILD/nvidia-kmod-169.04/_kmod_build_2.6.24-0.38.rc2.git6.fc9/usr/src/nv/Module.symvers -w -s
FATAL: modpost: GPL-incompatible module nvidia.ko uses GPL-only symbol 'pv_cpu_ops'
make[3]: *** [__modpost] Error 1
make[2]: *** [modules] Error 2
NVIDIA: left KBUILD.
nvidia.ko failed to build!
make[1]: *** [module] Error 1
make: *** [module] Error 2

Fedora development (rawhide 20071118)

see also : REGRESSION: 2.6.24 breaks nvidia and amd/ati binary drivers, by exporting paravirt symbols as GPL
http://www.mail-archive.com/linux-ke...msg229172.html

[zander]

This is a regression in the Linux kernel introduced by a change that split up the paravirt_ops dispatch table into a set of separate dispatch tables.

The NVIDIA Linux graphics driver implicitly picks up three dispatch tables, pv_cpu_ops, pv_mmu_ops and pv_irq_ops; the former two are only exported to GPL-licensed modules, the latter is exported to all modules.

More specifically, the Linux/x86 NVIDIA kernel module picks up pv_mmu_ops implicitely via pgd_val(), pte_val() and __flush_tlb(). It picks up pv_cpu_ops via local_irq_disable() (which translates to raw_local_irq_disable()), read_cr0(), write_cr0(), rdmsr() (which translates to paravirt_read_msr() and wrmsr() (which translates to paravirt_write_msr()).

This problem can not be worked around in the NVIDIA Linux graphics driver.

Due to changes to the layout of architecture specific include directories in recent Linux 2.6.24 pre-release kernels, the NVIDIA kernel module will fail to build against kernels built in separate output directories. This problem will be addressed in future driver releases.

[jml]

any ETA on a fix or workaround for this ?

Originally Posted by zander

This is a regression in the Linux kernel introduced by a change that split up the paravirt_ops dispatch table into a set of separate dispatch tables.

The NVIDIA Linux graphics driver implicitly picks up three dispatch tables, pv_cpu_ops, pv_mmu_ops and pv_irq_ops; the former two are only exported to GPL-licensed modules, the latter is exported to all modules.

More specifically, the Linux/x86 NVIDIA kernel module picks up pv_mmu_ops implicitely via pgd_val(), pte_val() and __flush_tlb(). It picks up pv_cpu_ops via local_irq_disable() (which translates to raw_local_irq_disable()), read_cr0(), write_cr0(), rdmsr() (which translates to paravirt_read_msr() and wrmsr() (which translates to paravirt_write_msr()).

This problem can not be worked around in the NVIDIA Linux graphics driver.

Due to changes to the layout of architecture specific include directories in recent Linux 2.6.24 pre-release kernels, the NVIDIA kernel module will fail to build against kernels built in separate output directories. This problem will be addressed in future driver releases.

[sangu]

Fixed in 169.07.

Thanks.

[Dragoran]

Originally Posted by sangu

Fixed in 169.07.

Thanks.

no its fixed in newer kernels (the symbols are exported for non GPL modules again).

[sangu]

Originally Posted by Dragoran

no its fixed in newer kernels (the symbols are exported for non GPL modules again).

$ uname -r
2.6.24-0.115.rc5.git5.fc9

$ dmesg | grep NVRM
NVRM: loading NVIDIA UNIX x86 Kernel Module 169.07 Thu Dec 13 18:42:56 PST 2007

$rpm -q --changelog kernel-2.6.24-0.115.rc5.git5.fc9 | more
* Thu Dec 20 2007 Dave Airlie <airlied@redhat.com>
- Update drm upstream patches and add basic r500 drm support

* Wed Dec 19 2007 Chuck Ebbert <cebbert@redhat.com>
- Enable CIFS upcall support.

* Wed Dec 19 2007 Kyle McMartin <kmcmartin@redhat.com>
- add --with sparse for people^Wsomeone who might care

* Wed Dec 19 2007 Kyle McMartin <kmcmartin@redhat.com>
- really disable sparse this time...

* Wed Dec 19 2007 Kyle McMartin <kmcmartin@redhat.com>
- 2.6.24-rc5-git5

And building nvidia kernel module works well in kernel-devel-2.6.24-0.121.rc6.fc9(2.6.24rc6).

[Dragoran]

yes thats what I said
note:
you are using 2.6.24-0.115.rc5.git5.fc9 which does not have the problem.
the thread was about: kernel-2.6.24-rc2-git6 which had this problem